Workflow of scATAC-seq with multiome - Normalization, Dimentionallity reduction and Harmony
2022-11-15
1 Introduction
In this report we will see how we do the normalization, the batach effect correction of multiome data, for each methodology.
2 Load packages
library(Signac)
library(Seurat)
library(ggplot2)
library(ggpubr)
library(tidyverse)
library(EnsDb.Hsapiens.v86)
library(plyr)
library(reshape2)
library(data.table)
library(GenomicRanges)
library(harmony)
library(hdf5r)
library(stringr)
library(ggpubr)
library(RColorBrewer)
library(magick)
library(knitr)
library(biovizBase)
library(patchwork)
set.seed(123)2.1 Directory path
# Paths
path_to_data <- here::here("results/R_objects/")
path_to_multiome_metadata <- here::here("data/tonsil_atlas_metadata_multiome.csv")
path_to_save <- here::here("results/R_objects/")2.2 Gene annotation
Extraction of gene annotations from EnsDb using hg38 as the reference assembly.
annotation <- GetGRangesFromEnsDb(ensdb = EnsDb.Hsapiens.v86)
seqlevelsStyle(annotation) <- "UCSC"
genome(annotation) <- "hg38"tonsil_merged <- readRDS(paste0(path_to_data,"1.tonsil_filtered_merged_all.rds"))3 Add metada
#become the first rowname as a column called "barcodes"
tonsil_merged@meta.data <- tibble::rownames_to_column(tonsil_merged@meta.data, "lib_name_barcode")
rn<-tonsil_merged@meta.data$lib_name_barcode
m1<-str_extract(rn, "BCLL\\_[\\d|\\d{2}]+\\_\\w\\_\\d")
tonsil_merged@meta.data$library_name<-m1
#Add rownames to metadata
rownames(tonsil_merged@meta.data)<-tonsil_merged@meta.data$lib_name_barcode4 Normalization and lineal and non-lineal dimentionallity reduction (LSI and UMAP)
4.1 scATAC
For normalization, Signac performs a two-step technique called TF-IDF (Term Frequency - Inverse Document Frequency). This technique normalizes the cells in order to correct possible differences in sequencing depth and normalize the peaks, increasing the signal of rare peaks. Here, we will use the RunTFIDF function.
Since scATAC-seq has a low signal/noise ratio is more difficult to select the top feature (peaks) as we do in scRNA-seq (genes). That is why we use the FindTopFeatures() command to select the top x% of the peaks for dimensionality reduction, or remove peaks that are present in less than x cells. In this case we will use all the min.cutoff=“q0” meaning that we will select the 100% of peaks.
Finally, for dimensionality reduction, we perform a mathematical technique called SVD or Singular Value Decomposition on the matrix returned by TD-IDF (SVD is also performed to generate PCAs) and using only the peaks selected in the previous step. We call LSI or latent semantic indexing the technique that combines a TF-IDF step followed by an SVD step, and it was successfully used for the first time in scATAC-seq analysis by Cusanovich et al., 2015.
We exclude the first dimension as this is typically correlated with sequencing depth Cells cluster completely separately in ATAC without harmony; so run harmony after SVD
RunSVD LSI
DefaultAssay(tonsil_merged) <- "ATAC"
tonsil_merged <- RunTFIDF(tonsil_merged)## Performing TF-IDF normalization## Warning in RunTFIDF.default(object = GetAssayData(object = object, slot =
## "counts"), : Some features contain 0 total countstonsil_merged <- FindTopFeatures(tonsil_merged, min.cutoff = "q0")
tonsil_merged <- RunSVD(tonsil_merged)## Running SVD## Scaling cell embeddings4.1.1 Plot the Depth correlation plot
Compute the correlation between total counts and each reduced dimension component.
LSI component is typically highly correlated with sequencing depth. The first LSI component often captures sequencing depth (technical variation) rather than biological variation. If this is the case, the component should be removed from downstream analysis. We can assess the correlation between each LSI component and sequencing depth using the DepthCor() function:
For scRNA-seq data we don’t typically observe such a strong relationship between the first PC and sequencing depth, and so usually retain the first PC in downstream analyses.
DepthCor(tonsil_merged)
Here we see there is a very strong correlation between the first LSI component and the totalnumber of counts for the cell, so we will perform downstream steps without this component.
4.1.2 UMAP representation
- dimensional reduction key, specifies the string before the number for the dimension names. UMAP by default
- reduction.name: Name to store dimensional reduction under in the Seurat object
tonsil_merged <- RunUMAP(
tonsil_merged,
dims = 2:40,
reduction = "lsi",
reduction.name = "umap.atac",
reduction.key = "atacUMAP_"
)## Warning: The default method for RunUMAP has changed from calling Python UMAP via reticulate to the R-native UWOT using the cosine metric
## To use Python UMAP via reticulate, set umap.method to 'umap-learn' and metric to 'correlation'
## This message will be shown once per session## 21:35:29 UMAP embedding parameters a = 0.9922 b = 1.112## 21:35:29 Read 11954 rows and found 39 numeric columns## 21:35:29 Using Annoy for neighbor search, n_neighbors = 30## 21:35:29 Building Annoy index with metric = cosine, n_trees = 50## 0% 10 20 30 40 50 60 70 80 90 100%## [----|----|----|----|----|----|----|----|----|----|## **************************************************|
## 21:35:30 Writing NN index file to temp file /var/folders/kz/y10np0cj213fybqz181z9lghbzr42p/T//RtmpS9FgdI/file306a77bbee6
## 21:35:30 Searching Annoy index using 1 thread, search_k = 3000
## 21:35:33 Annoy recall = 100%
## 21:35:36 Commencing smooth kNN distance calibration using 1 thread with target n_neighbors = 30
## 21:35:38 Initializing from normalized Laplacian + noise (using irlba)
## 21:35:39 Commencing optimization for 200 epochs, with 491382 positive edges
## 21:35:46 Optimization finishedatac.umap<-DimPlot(
tonsil_merged,
reduction = "umap.atac",
group.by = "library_name",
pt.size = 0.1
) + ggtitle('scATAC UMAP') + NoLegend()
atac.umap
#split_by: library ,edad, generoatac.umap 
4.2 scRNA-seq
Our aim is to detect and exclude empty droplets or deth cells (lysed cells). Lysed cells have 3 hallmarks: - (1) low library size (total UMI), - (2) low library complexity (number of detected genes) a - (3) high fraction of mitochondrial expression (cytosolic mRNA leaks out of the cell).
4.2.1 Normalization and linear dimensional reduction-
4.2.2 NormalizeData (Log Normalization)
DefaultAssay(tonsil_merged) <- "RNA"
tonsil_merged <- NormalizeData(
tonsil_merged,
normalization.method = "LogNormalize",
scale.factor = 1e4
)
tonsil_merged <- tonsil_merged %>%
FindVariableFeatures(nfeatures = 3000) %>%
ScaleData() %>%
RunPCA() PCAPlot(tonsil_merged,
group.by = "library_name")
ElbowPlot(object = tonsil_merged)
4.2.3 UMAP representation
tonsil_merged <- RunUMAP(
tonsil_merged,
dims = 1:30,
reduction = "pca",
reduction.name = "umap.rna",
reduction.key = "rnaUMAP_"
)## 21:36:13 UMAP embedding parameters a = 0.9922 b = 1.112## 21:36:13 Read 11954 rows and found 30 numeric columns## 21:36:13 Using Annoy for neighbor search, n_neighbors = 30## 21:36:13 Building Annoy index with metric = cosine, n_trees = 50## 0% 10 20 30 40 50 60 70 80 90 100%## [----|----|----|----|----|----|----|----|----|----|## **************************************************|
## 21:36:14 Writing NN index file to temp file /var/folders/kz/y10np0cj213fybqz181z9lghbzr42p/T//RtmpS9FgdI/file306a5c447809
## 21:36:14 Searching Annoy index using 1 thread, search_k = 3000
## 21:36:17 Annoy recall = 100%
## 21:36:18 Commencing smooth kNN distance calibration using 1 thread with target n_neighbors = 30
## 21:36:20 Initializing from normalized Laplacian + noise (using irlba)
## 21:36:20 Commencing optimization for 200 epochs, with 508910 positive edges
## 21:36:28 Optimization finishedrna.umap<-DimPlot(
tonsil_merged,
reduction = "umap.rna",
group.by = "library_name",
pt.size = 0.1) + NoLegend() + ggtitle('scRNA UMAP')
rna.umap
plot(rna.umap)
atac.umap + rna.umap
5 Harmony Integration (correct data by bath effect)
Pass the Seurat object to the RunHarmony function and specify which variable to integrate out. A Seurat object is returned with corrected Harmony coordinates.
5.1 scATAC
DefaultAssay(tonsil_merged) <- "ATAC"
tonsil_merged <- RunHarmony(
object = tonsil_merged,
reduction = "lsi",
dims = 2:40,
group.by.vars = "library_name",
assay.use = "ATAC",
project.dim = FALSE,
reduction.save = "harmony_atac"
)5.1.1 UMAP representation
tonsil_merged <- RunUMAP(
tonsil_merged,
dims = 2:40,
reduction = "harmony_atac",
reduction.name = "umap.atac",
reduction.key = "atacUMAP_"
)## 21:36:53 UMAP embedding parameters a = 0.9922 b = 1.112## 21:36:53 Read 11954 rows and found 39 numeric columns## 21:36:53 Using Annoy for neighbor search, n_neighbors = 30## 21:36:53 Building Annoy index with metric = cosine, n_trees = 50## 0% 10 20 30 40 50 60 70 80 90 100%## [----|----|----|----|----|----|----|----|----|----|## **************************************************|
## 21:36:54 Writing NN index file to temp file /var/folders/kz/y10np0cj213fybqz181z9lghbzr42p/T//RtmpS9FgdI/file306a6b8aa574
## 21:36:54 Searching Annoy index using 1 thread, search_k = 3000
## 21:36:57 Annoy recall = 100%
## 21:36:58 Commencing smooth kNN distance calibration using 1 thread with target n_neighbors = 30
## 21:37:00 Initializing from normalized Laplacian + noise (using irlba)
## 21:37:00 Commencing optimization for 200 epochs, with 498624 positive edges
## 21:37:08 Optimization finishedHarm_peak<-DimPlot(
tonsil_merged,
reduction = "umap.atac",
group.by = "library_name",
pt.size = 0.1
) + NoLegend() + ggtitle('Peak Harmony')6 scRNA
harmony in RNA-seq
tonsil_merged <- RunUMAP(
tonsil_merged,
dims = 2:40,
reduction = "harmony_rna",
reduction.name = "umap.rna",
reduction.key = "rnaUMAP_"
)## 21:37:21 UMAP embedding parameters a = 0.9922 b = 1.112## 21:37:21 Read 11954 rows and found 39 numeric columns## 21:37:21 Using Annoy for neighbor search, n_neighbors = 30## 21:37:21 Building Annoy index with metric = cosine, n_trees = 50## 0% 10 20 30 40 50 60 70 80 90 100%## [----|----|----|----|----|----|----|----|----|----|## **************************************************|
## 21:37:22 Writing NN index file to temp file /var/folders/kz/y10np0cj213fybqz181z9lghbzr42p/T//RtmpS9FgdI/file306a66de7cdc
## 21:37:22 Searching Annoy index using 1 thread, search_k = 3000
## 21:37:25 Annoy recall = 100%
## 21:37:26 Commencing smooth kNN distance calibration using 1 thread with target n_neighbors = 30
## 21:37:28 Initializing from normalized Laplacian + noise (using irlba)
## 21:37:28 Commencing optimization for 200 epochs, with 512340 positive edges
## 21:37:36 Optimization finishedHarm_rna<-DimPlot(
tonsil_merged,
reduction = "umap.rna",
group.by = "library_name",
pt.size = 0.1
) + NoLegend() + ggtitle('RNA Harmony')6.1 scATAC and RNAseq Harmony
Harm_peak+Harm_rna + plot_annotation(title = 'Harmony ATAC and RNA UMAP visualization')
7 Save
saveRDS(tonsil_merged,paste0(path_to_save,"2.tonsil_merged_harmony.rds"))Session Info
sessionInfo()## R version 4.2.1 (2022-06-23)
## Platform: x86_64-apple-darwin17.0 (64-bit)
## Running under: macOS Big Sur ... 10.16
##
## Matrix products: default
## BLAS: /Library/Frameworks/R.framework/Versions/4.2/Resources/lib/libRblas.0.dylib
## LAPACK: /Library/Frameworks/R.framework/Versions/4.2/Resources/lib/libRlapack.dylib
##
## locale:
## [1] es_ES.UTF-8/es_ES.UTF-8/es_ES.UTF-8/C/es_ES.UTF-8/es_ES.UTF-8
##
## attached base packages:
## [1] stats4 stats graphics grDevices utils datasets methods
## [8] base
##
## other attached packages:
## [1] patchwork_1.1.2 biovizBase_1.44.0
## [3] knitr_1.40 magick_2.7.3
## [5] RColorBrewer_1.1-3 hdf5r_1.3.7
## [7] harmony_0.1.0 Rcpp_1.0.9
## [9] data.table_1.14.4 reshape2_1.4.4
## [11] plyr_1.8.7 EnsDb.Hsapiens.v86_2.99.0
## [13] ensembldb_2.20.2 AnnotationFilter_1.20.0
## [15] GenomicFeatures_1.48.4 AnnotationDbi_1.60.0
## [17] Biobase_2.58.0 GenomicRanges_1.50.1
## [19] GenomeInfoDb_1.34.2 IRanges_2.32.0
## [21] S4Vectors_0.36.0 BiocGenerics_0.44.0
## [23] forcats_0.5.2 stringr_1.4.1
## [25] dplyr_1.0.10 purrr_0.3.5
## [27] readr_2.1.3 tidyr_1.2.1
## [29] tibble_3.1.8 tidyverse_1.3.2
## [31] ggpubr_0.4.0 ggplot2_3.3.6
## [33] sp_1.5-0 SeuratObject_4.1.2
## [35] Seurat_4.2.0 Signac_1.8.0
## [37] BiocStyle_2.24.0
##
## loaded via a namespace (and not attached):
## [1] rappdirs_0.3.3 rtracklayer_1.56.1
## [3] scattermore_0.8 bit64_4.0.5
## [5] irlba_2.3.5.1 DelayedArray_0.24.0
## [7] rpart_4.1.19 KEGGREST_1.38.0
## [9] RCurl_1.98-1.9 generics_0.1.3
## [11] cowplot_1.1.1 RSQLite_2.2.18
## [13] RANN_2.6.1 future_1.28.0
## [15] bit_4.0.4 tzdb_0.3.0
## [17] spatstat.data_3.0-0 xml2_1.3.3
## [19] lubridate_1.8.0 httpuv_1.6.6
## [21] SummarizedExperiment_1.28.0 assertthat_0.2.1
## [23] gargle_1.2.1 xfun_0.34
## [25] hms_1.1.2 jquerylib_0.1.4
## [27] evaluate_0.17 promises_1.2.0.1
## [29] fansi_1.0.3 restfulr_0.0.15
## [31] progress_1.2.2 dbplyr_2.2.1
## [33] readxl_1.4.1 igraph_1.3.5
## [35] DBI_1.1.3 htmlwidgets_1.5.4
## [37] spatstat.geom_2.4-0 googledrive_2.0.0
## [39] ellipsis_0.3.2 backports_1.4.1
## [41] bookdown_0.29 biomaRt_2.52.0
## [43] deldir_1.0-6 MatrixGenerics_1.10.0
## [45] vctrs_0.5.0 here_1.0.1
## [47] ROCR_1.0-11 abind_1.4-5
## [49] cachem_1.0.6 withr_2.5.0
## [51] BSgenome_1.64.0 progressr_0.11.0
## [53] checkmate_2.1.0 sctransform_0.3.5
## [55] GenomicAlignments_1.32.1 prettyunits_1.1.1
## [57] goftest_1.2-3 cluster_2.1.4
## [59] lazyeval_0.2.2 crayon_1.5.2
## [61] labeling_0.4.2 pkgconfig_2.0.3
## [63] nlme_3.1-160 ProtGenerics_1.28.0
## [65] nnet_7.3-18 rlang_1.0.6
## [67] globals_0.16.1 lifecycle_1.0.3
## [69] miniUI_0.1.1.1 filelock_1.0.2
## [71] BiocFileCache_2.6.0 modelr_0.1.9
## [73] dichromat_2.0-0.1 rprojroot_2.0.3
## [75] cellranger_1.1.0 polyclip_1.10-4
## [77] matrixStats_0.62.0 lmtest_0.9-40
## [79] Matrix_1.5-1 carData_3.0-5
## [81] zoo_1.8-11 reprex_2.0.2
## [83] base64enc_0.1-3 ggridges_0.5.4
## [85] googlesheets4_1.0.1 png_0.1-7
## [87] viridisLite_0.4.1 rjson_0.2.21
## [89] bitops_1.0-7 KernSmooth_2.23-20
## [91] Biostrings_2.66.0 blob_1.2.3
## [93] parallelly_1.32.1 spatstat.random_2.2-0
## [95] jpeg_0.1-9 rstatix_0.7.0
## [97] ggsignif_0.6.4 scales_1.2.1
## [99] memoise_2.0.1 magrittr_2.0.3
## [101] ica_1.0-3 zlibbioc_1.44.0
## [103] compiler_4.2.1 BiocIO_1.6.0
## [105] fitdistrplus_1.1-8 Rsamtools_2.12.0
## [107] cli_3.4.1 XVector_0.38.0
## [109] listenv_0.8.0 pbapply_1.5-0
## [111] htmlTable_2.4.1 Formula_1.2-4
## [113] MASS_7.3-58.1 mgcv_1.8-41
## [115] tidyselect_1.2.0 stringi_1.7.8
## [117] highr_0.9 yaml_2.3.6
## [119] latticeExtra_0.6-30 ggrepel_0.9.1
## [121] grid_4.2.1 VariantAnnotation_1.42.1
## [123] sass_0.4.2 fastmatch_1.1-3
## [125] tools_4.2.1 future.apply_1.9.1
## [127] parallel_4.2.1 rstudioapi_0.14
## [129] foreign_0.8-83 gridExtra_2.3
## [131] farver_2.1.1 Rtsne_0.16
## [133] digest_0.6.30 BiocManager_1.30.19
## [135] rgeos_0.5-9 shiny_1.7.3
## [137] car_3.1-1 broom_1.0.1
## [139] later_1.3.0 RcppAnnoy_0.0.19
## [141] httr_1.4.4 colorspace_2.0-3
## [143] rvest_1.0.3 XML_3.99-0.11
## [145] fs_1.5.2 tensor_1.5
## [147] reticulate_1.26 splines_4.2.1
## [149] uwot_0.1.14 RcppRoll_0.3.0
## [151] spatstat.utils_3.0-1 plotly_4.10.0
## [153] xtable_1.8-4 jsonlite_1.8.3
## [155] R6_2.5.1 Hmisc_4.7-1
## [157] pillar_1.8.1 htmltools_0.5.3
## [159] mime_0.12 glue_1.6.2
## [161] fastmap_1.1.0 BiocParallel_1.30.4
## [163] codetools_0.2-18 utf8_1.2.2
## [165] lattice_0.20-45 bslib_0.4.1
## [167] spatstat.sparse_3.0-0 curl_4.3.3
## [169] leiden_0.4.3 interp_1.1-3
## [171] survival_3.4-0 rmarkdown_2.17
## [173] munsell_0.5.0 GenomeInfoDbData_1.2.9
## [175] haven_2.5.1 gtable_0.3.1
## [177] spatstat.core_2.4-4